Exploring variables optimization methods to screen surface-enhanced Raman spectroscopy characteristic peaks for rapid detection of difenoconazole pesticides in tea

Abstract

Detection of pesticide residues in tea is a requisite technique to safeguard consumer food safety. This paper carried out rapid quantitative analysis of difenoconazole residues by surface-enhanced Raman spectroscopy (SERS) combined with chemometric methods. Primary secondary amine (PSA), nano bamboo charcoal (NBC) and anhydrous MgSO4 were employed to eliminate the impact of tea matrix in tea on the difenoconazole pesticides investigation. SERS spectral data excited with gold colloid was collected from 97 tea samples. Competitive adaptive reweighted sampling (CARS) was employed to choose some optimal spectral variables, which were coincident with experiment characteristic peaks and theoretical calculation. Then, random forest (RF) was adapted to develop quantitative analysis models for monitoring difenoconazole residues, and compared with partial least squares (PLS) models. The results demonstrated that the prediction performance of the models using spectra variables subset screened by CARS was significantly improved. In addition, non-linear RF regression models outperformed linear PLS regression models in predicting difenoconazole concentration in tea. The best results of the CARS_RF model were obtained with correlation coefficient (R), root mean square error (RMSE) and ratio performance deviation (RPD) in prediction set of 0.97, 2.50 and 4.01, respectively. The overall results encouraged the development of a rapid, simple and convenient spectral spectroscopic quantitative analysis method of pesticide residues in tea.